The Assimilation of Envisat data (ASSET) project

HAL Id: hal-00328494

https://hal.archives-ouvertes.fr/hal-00328494

Submitted on 11 Apr 2007

HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.

L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.

W. A. Lahoz, A. J. Geer, Slimane Bekki, N. Bormann, S. Ceccherini, H.

Elbern, Q. Errera, H. J. Eskes, D. Fonteyn, D. R. Jackson, et al.

To cite this version:

W. A. Lahoz, A. J. Geer, Slimane Bekki, N. Bormann, S. Ceccherini, et al.. The Assimilation of

Envisat data (ASSET) project. Atmospheric Chemistry and Physics, European Geosciences Union,

2007, 7 (7), pp.1773-1796. �10.5194/acp-7-1773-2007�. �hal-00328494�

www.atmos-chem-phys.net/7/1773/2007/

© Author(s) 2007. This work is licensed under a Creative Commons License.

Chemistry and Physics

The Assimilation of Envisat data (ASSET) project

W. A. Lahoz ¹ , A. J. Geer ^1,* , S. Bekki ² , N. Bormann ³ , S. Ceccherini ⁴ , H. Elbern ⁵ , Q. Errera ⁶ , H. J. Eskes ⁷ ,

D. Fonteyn ⁶ , D. R. Jackson ⁸ , B. Khattatov ⁹ , M. Marchand ² , S. Massart ¹⁰ , V.-H. Peuch ¹¹ , S. Rharmili ² , M. Ridolfi ¹² , A. Segers ⁷ , O. Talagrand ¹³ , H. E. Thornton ⁸ , A. F. Vik ¹⁴ , and T. von Clarmann ¹⁵

1 Data Assimilation Research Centre, University of Reading, Reading, UK

2 Service Aeronomie, Universite Pierre et Marie Curie, Paris, France

3 ECMWF, Reading, UK

4 CNR-IFAC, Florence, Italy

5 RIU, University of K¨oln, K¨oln, Germany

6 BIRA-IASB, Brussels, Belgium

7 KNMI, De Bilt, The Netherlands

8 Met Office, Exeter, UK

9 Fusion Numerics, Boulder, CO, USA

10 CERFACS, Toulouse, France

11 M´et´eo-France, Toulouse, France

12 University of Bologna, Bologna, Italy

13 LMD, Paris, France

14 NILU, Kjeller, Norway

15 Forschungszentrum Karlsruhe, IMK, Germany

* now at: ECMWF, Reading, UK

Received: 4 October 2006 – Published in Atmos. Chem. Phys. Discuss.: 8 December 2006 Revised: 27 February 2007 – Accepted: 27 March 2007 – Published: 11 April 2007

Abstract. This paper discusses the highlights of the EU- funded “Assimilation of Envisat data” (ASSET) project, which has involved assimilation of Envisat atmospheric con- stituent and temperature data into systems based on Numer- ical Weather Prediction (NWP) models and chemical trans- port models (CTMs). Envisat was launched in 2002 and is one of the largest Earth Observation (EO) satellites ever built.

It carries several sophisticated EO instruments providing in- sights into chemistry and dynamics of the atmosphere. In this paper we focus on the assimilation of temperature and constituents from Envisat.

The overarching theme of the ASSET project has been to bring together experts from all aspects of the data assimila- tion problem. This has allowed ASSET to address several themes comprehensively: enhancement of NWP analyses by assimilation of research satellite data; studies of the distri- bution of stratospheric chemical species by assimilation of research satellite data into CTM systems; objective assess- ment of the quality of ozone analyses; studies of the spatial and temporal evolution of tropospheric pollutants; enhanced retrievals of Envisat data; and data archival and dissemina- tion.

Correspondence to: W. A. Lahoz (swslahoz@rdg.ac.uk)

Among the results from the ASSET project, many of which are firsts in their field, we can mention: a positive im- pact on NWP analyses from assimilation of height-resolved stratospheric humidity and temperature data, and assimila- tion of limb radiances; the extraction of temperature infor- mation from the assimilation of chemical species into CTMs;

a first intercomparison between ozone assimilation systems;

the extraction of information on tropospheric pollution from assimilation of Envisat data; and the large potential of the En- visat MIPAS dataset. This paper discusses these, often novel, developments and results. Finally, achievements of, and rec- ommendations from, the ASSET project are presented.

1 Introduction

The EU-funded “Assimilation of Envisat data” (ASSET)

project (http://darc.nerc.ac.uk/asset) is a major European ini-

tiative in Earth Observation (EO); it has run for the period

January 2003–June 2006. Its overall rationale is to use the

techniques of data assimilation, DA (e.g. Kalnay, 2003) to

develop a European capability for chemical and UV fore-

casting, and provide analyses for coupled climate/chemistry

studies. The objectives of the ASSET project are: (1) assess

the strategies for exploiting research satellite data by the Nu- merical Weather Prediction (NWP) community, and (2) us- ing this data to investigate the distribution and variability of atmospheric chemical species. To address these objectives, the ASSET project brought together experts from all aspects of the DA problem: NWP systems; chemical DA; numeri- cal modelling; meteorology; retrieval theory; DA theory; EO measurements; data analysis; and data management. In this paper we focus on the assimilation of temperature and con- stituent data from Envisat. The Envisat satellite carries on board ten sensors. It flies in a sun-synchronous polar orbit of ∼800 km altitude. It has a 10:00 a.m. mean solar time for the descending node, and an orbit inclination of 98.55 ^◦ . The repeat cycle is 35 days, and because most sensors are wide swath, it provides a complete coverage of the globe within 1–3 days.

The ASSET partners brought to the project experience with different DA systems, DA techniques and assimilated data (see http://darc.nerc.ac.uk/asset). The DA systems used by the ASSET partners to assimilate constituent data from Envisat fell into three categories: NWP DA systems based on dynamical models, often General Circulation Models (GCMs); chemical DA systems based on Chemical Trans- port Models (CTMs) driven by off-line NWP analyses or short-term forecasts; coupled DA systems (usually a GCM coupled with a CTM). In this paper, we focus on the first two systems. The DA techniques used in the ASSET project included three- and four-dimensional variational data assim- ilation (3D-Var and 4D-Var, respectively), and Kalman Filter (KF) methods (Kalnay, 2003). Another variational approach used in the ASSET project was 3D-FGAT (First Guess at the Appropriate Time), a variant of 3D-Var. The Envisat data as- similated into the DA systems involved in ASSET consisted of level 1 data (radiances or line densities) and level 2 data (height-resolved or total-column retrievals) from the follow- ing instruments: AATSR (Advanced Along Track Scanning Radiometer), GOMOS (Global Ozone Monitoring by Occul- tation of Stars), MIPAS (Michelson Interferometer for Pas- sive Atmospheric Sounding), and SCIAMACHY (SCanning Imaging Absorption spectroMeter for Atmospheric CHartog- raphY). We mainly use data from MIPAS, chiefly because at the time when the work described was done (2005), this dataset was, in general, more mature than that from SCIA- MACHY or GOMOS.

MIPAS measures the atmospheric limb emission spec- trum in the frequency interval 680–2410 cm ⁻¹ over the al- titude range 6–68 km. The spacing between measurements is ∼3 km through the stratosphere, but larger above. MIPAS typically samples a volume ∼3 km in the vertical, ∼30 km in azimuth and ∼300 km along the line of sight. Precise val- ues will differ slightly with species and between retrievals:

the true vertical resolution is described by an averaging ker- nel (Rodgers, 2000). The MIPAS measured spectra are anal- ysed by the ESA ground processor to provide height-resolved profiles of pressure, temperature, and six key atmospheric

species (known as “target species”): H ₂ O, O ₃ , HNO ₃ , CH ₄ , N ₂ O and NO ₂ . Of these MIPAS measurements, humidity, temperature (Sect. 2.1) and ozone (Sect. 2.3) have been as- similated into NWP and CTM systems, and several tropo- spheric constituents, including ozone, HNO ₃ and NO ₂ have been assimilated into a CTM system (Sect. 2.4). Details of the SCIAMACHY datasets used in the ASSET project can be found in Eskes et al. (2005) and Segers et al. (2005b). De- tails of the GOMOS datasets used in the ASSET project can be found in Marchand et al. (2004).

In this paper we describe the assimilation of both level 1 and level 2 data from Envisat. Note that nowadays, level 1 data are routinely assimilated by NWP agencies (e.g. Saun- ders et al., 1999; McNally et al., 2006). Besides the work described in this paper, there have been considerable ef- forts to evaluate Envisat data and use it for scientific stud- ies. Worthy of mention are two special issues in At- mos. Chem. Phys. on the evaluation of Envisat data: (i) Geophysical Validation of SCIAMACHY 2002–2004 (Eds.

Kelder, Platt and Simon), http://www.atmos-chem-phys.net/

special issue19.html (2005); and (ii) MIPAS (Michelson In- terferometer for Passive Atmospheric Sounding): Potential of the experiment, data processing and validation of results (Eds. Espy and Hartogh), http://www.atmos-chem-phys.net/

special issue70.html (2006). Data from the atmospheric chemistry instruments aboard Envisat have also been used to study the unprecedented Antarctic ozone hole split of September 2002 (special issue in J. Atmos. Sci., 2005, vol. 62).

This paper unifies in one publication selected highlights of the ASSET project. Some of the work in the ASSET project is described elsewhere in the literature in more de- tail, but this paper also includes much work that will not be published separately, and all of the results are explained in a wider context. Section 2 describes ASSET project high- lights. Section 3 summarizes achievements and results of the ASSET project (many of them firsts in their field), and presents recommendations.

2 ASSET project highlights

In Sect. 2.1 we discuss the impacts of assimilation of height-

resolved humidity and temperature data from MIPAS, and

of direct assimilation of limb radiances from MIPAS. In

Sect. 2.2 we discuss the stratospheric distribution of NO _x

and NO _y (NO _x =NO+NO ₂ ; NO _y = NO _x +HNO ₃ plus other re-

lated chemical species), and the extraction of temperature in-

formation from constituent information, using analyses de-

rived from assimilation of GOMOS data. In Sect. 2.3 we

objectively evaluate the assimilation of MIPAS and SCIA-

MACHY ozone data. Sections 2.1–2.3 focus mainly on the

stratosphere, but in Sect. 2.4, the assimilation of tropospheric

constituents from Envisat is discussed. Section 2.5 discusses

enhanced retrievals of MIPAS data. Section 2.6 discusses

Table 1. List of DA systems participating in the ASSIC project (Sect. 2.3). The Met Office, ECMWF MIPAS and BASCOE v3q33 systems were also used to study the assimilation of humidity and temperature, and the assimilation of limb radiances (Sect. 2.1). See Geer et al. (2006, 2007) for more details of the DA systems. SBUV is the Solar Backscatter UV radiometer; GOME is the Global Ozone Monitoring Experiment.

Name Type Winds Scheme Ozone observations Ozone photochemistry Heterogeneous

ozone chemistry ECMWF operational NWP GCM 4D-Var SBUV, GOME total columns, MIPAS

v4.59 from 07/10/03

Cariolle v1.2 T<195 K term ECMWF MIPAS NWP GCM 4D-Var As ECMWF operational, but through-

out ASSIC period

Cariolle v1.2 T<195 K term

DARC/Met Office NWP GCM 3D-Var MIPAS v4.61 Cariolle v1.0 Cold tracer

KNMI TEMIS CTM ECMWF Sub-optimal KF SCIAMACHY TOSOMI total columns (Eskes et al., 2005)

LINOZ Cold tracer

KNMI SCIAMACHY profiles CTM ECMWF Sub-optimal KF SCIAMACHY profiles, IFE Bremen v1.6

Cariolle v1.0 Cold tracer BASCOE v3d24

(BIRA-IASB)

CTM ECMWF 4D-Var MIPAS v4.61 57 species. Mesospheric

chemistry

PSCBox BASCOE v3q33

(BIRA-IASB)

CTM ECMWF 4D-Var MIPAS v4.61 As BASCOE v3d24 PSC parametrization

MOCAGE-PALM/Cariolle (M´et´eo-France/CERFACS)

CTM Arpege 3D-FGAT MIPAS v4.61 Cariolle v2.1 T<195 K term

MOCAGE-PALM/Reprobus (M´et´eo-France/CERFACS)

CTM Arpege 3D-FGAT MIPAS v4.61 Reprobus (Lef`evre et al.,

1994). Tropospheric chemistry

Carslaw et al. (1995)

MIMOSA

(Service d’Aeronomie)

CTM ECMWF Sub-optimal KF MIPAS v4.61 None None

Juckes (2006) (RAL)

CTM ECMWF Direct inversion MIPAS v4.61 None None

how the data produced by the ASSET project is archived and disseminated. Section 3 brings together these results, pro- vides conclusions and identifies how ASSET is greater than the sum of its parts.

Table 1 provides a list of the DA systems participat- ing in the ASSET ozone intercomparison (ASSIC) project (Sect. 2.3; Geer et al., 2006). Geer et al. (2007) provides fur- ther details of the ozone photochemistry and heterogeneous parametrizations used in the ASSIC project. Some of these DA systems were also used to study the assimilation of hu- midity and temperature, and the assimilation of limb radi- ances (Sect. 2.1).

2.1 Assimilation of MIPAS data

2.1.1 Assimilation of humidity retrievals

The daily variability of the water vapour field in the strato- sphere is poorly known. Gaining an improved knowledge of this variability is very desirable, as upper troposphere/lower stratosphere (UTLS) water vapour is known to play an im- portant role in many aspects of meteorology, including ra- diation, dynamics, chemistry and climate change (SPARC, 2000). The operational assimilation of humidity data in the stratosphere by NWP centres is limited by the availability of suitable data (Simmons et al., 1999). The reduction in specific humidity by four to five orders of magnitude from the surface to 0.1 hPa, also makes the assimilation problem considerably more difficult. Due to assimilation problems,

both the Met Office (N. B. Ingleby, personal communication, 2003) and ECMWF (H´olm et al., 2002) have made to date ad hoc fixes to constrain the stratospheric humidity field. How- ever, with the availability of high quality humidity data from Envisat, with high spatial and temporal density, it is now ap- propriate to revisit the issue of stratospheric humidity assim- ilation. Here, we present preliminary results from the assim- ilation of MIPAS humidity profiles into three different DA systems. This is the first study to perform such a compari- son.

In this section we discuss the assimilation of MIPAS hu- midity profiles from the ESA ground processor (ESA, 2004) – in Sect. 2.5 we discuss the production of MIPAS data from outside ESA. The ESA MIPAS humidity profiles are avail- able from 12–60 km. The estimated error standard deviation for v4.61 is 10–20% near the tropopause, 3–10% in the 15–

40 km layer, and 10–20% in the 40–60 km layer; the total

error (random plus systematic, i.e., bias) is 20–25% near the

tropopause, 15–20% in the 15–40 km layer, and 20–50% in

the 40–60 km layer (Raspollini et al., 2006). The main sys-

tematic errors are associated with spectroscopy and neglect

of horizontal temperature variability in the retrievals. Com-

parison of the MIPAS humidity profiles with balloon and

aircraft data (Oelhaf et al., 2004), ground-based radiome-

ter and lidar data (Pappalardo et al., 2004) and HALOE

(Halogen Occultation Experiment), SAGE II (Statospheric

Aerosol and Gas Experiment) and POAM III (Polar Ozone

and Aerosol Measurement) satellite data (Weber et al., 2004)

shows good agreement between 15 km and 30 km. However,

above 30 km the MIPAS retrievals have a positive bias of up to 20% compared to the satellite data and of 7–15% com- pared to ground-based radiometer and lidar data. In the UTLS region the MIPAS retrievals have a small negative bias compared to the balloon and aircraft data.

Three different groups participating in the ASSET project have assimilated MIPAS water vapour profile data (v4.61):

the Met Office, ECMWF and BASCOE (BIRA-IASB). Two are NWP systems (Met Office and ECMWF); BASCOE as- similates data into a CTM. Their DA systems are summa- rized in Table 1. Because of the large number of variables involved, variational data assimilation schemes do not per- form the minimization in the model space but, instead, use a transformed or control space. The elements of this control space are termed the control variables (Kalnay, 2003). The discussion below describes what control variables BASCOE, Met Office and ECMWF use when assimilating stratospheric humidity.

The BASCOE DA system is only concerned with strato- spheric humidity analyses and hence can use water vapour mixing ratio as the control variable, whereas the Met Of- fice, as well as ECMWF, perform a combined stratospheric and tropospheric humidity analysis. In these experiments, the Met Office uses normalized relative humidity as the hu- midity control variable throughout the atmosphere. ECMWF uses normalized specific humidity in the stratosphere, and normalized relative humidity in the troposphere. The con- cept of normalization to make the background errors (i.e., the short-term forecast errors in the data assimilation cycle) flow dependent and Gaussian was formulated by H´olm et al. (2002). Dee and da Silva (2003) highlighted the nega- tive impact of temperature observations for humidity assim- ilation and, following their recommendations, any correla- tion between temperature and specific humidity is removed from the control variable in the stratosphere of both the Met Office and ECMWF systems. This is done by using a rela- tive humidity-like control variable which is calculated using background (i.e., short-term forecast), rather than analysed, temperatures. The Met Office has tested other formulations of the control variable, and these results appear later in this sub-section.

The Met Office and ECMWF generated their background error covariance matrices using the NMC method (Parrish and Derber, 1992) and ensemble method (Fisher, 2003), re- spectively. The BASCOE background error covariances are much simpler, with no vertical or horizontal error correla- tions and standard deviations equal to 20% of the background humidity field. Although there are no error correlations in the BASCOE system, information from MIPAS observations is still spread via the observation operator, which transforms variables from model space to observation space. The BAS- COE observation operator averages the information from the eight grid points surrounding the observation point; the rela- tively coarse grid resolution in the BASCOE CTM also helps to spread this information.

For both ECMWF and the Met Office, the assimilation schemes used are experimental and are different from the op- erational schemes, where humidity assimilation in the strato- sphere is respectively switched off or constrained. The Met Office and ECMWF systems parametrize both water vapour production by methane oxidation in the stratosphere and wa- ter vapour loss through photolysis in the mesosphere. In the BASCOE system, these processes are modelled explicitly.

The humidity analyses from all three groups have vary- ing accuracies that depend on the DA system, the level and the latitude. Figure 1 shows the monthly mean zonal water vapour analyses for September 2003 for the ECMWF and BASCOE systems. The analyses were interpolated onto a common grid to enable a simpler comparison. The common grid is the same as that used in the ASSIC project (Sect. 2.3;

Geer et al., 2006): 3.75 ^◦ longitude × 2.5 ^◦ latitude, with 37 fixed pressure levels.

The monthly mean analyses show good agreement with the UARS reference atmosphere for September (http://code916.gsfc.nasa.gov/Public/Analysis/UARS/urap/

home.html) (not shown). A number of well-known features can be seen in the stratospheric analyses from BASCOE and ECMWF (we postpone discussion of the Met Office analyses until later in this sub-section). These include the very dry tropical tropopause (near 100 hPa) and the dehydration within the Antarctic winter polar vortex (between 100 hPa and 50 hPa; SPARC, 2000). The stratospheric presence of a layer of dry (∼3 ppmv) air around the 100 to 200 hPa layer is indicative that some of the dry air coming into the stratosphere in the tropics is quickly transported toward the Pole at these levels. There is also slow upward transport of dry air at low latitudes via the Brewer-Dobson circula- tion. As the air is transported upwards, methane oxidation leads to an increase in humidity, which is reflected in the relatively moist air seen in the upper stratosphere and lower mesosphere (levels above 10 hPa). Near the stratopause (near 1 hPa) there is an overturning of the stratospheric air because of a change in the pattern of the Brewer-Dobson circulation. The upward low latitude transport is replaced by poleward transport, and associated downward transport at high latitudes. Thus, at high latitudes there is downward transport of the moist air from the upper stratosphere/lower mesosphere to the mid stratosphere, most especially in the winter high latitudes, where this downward transport is stronger.

Between the tropopause and 1 hPa, the zonal mean monthly analyses for the BASCOE and ECMWF systems are reasonably similar. The BASCOE analyses show a drier UTLS region at most latitudes, whereas the ECMWF anal- yses show a more distinct dry tropical tropopause region.

Consequently, the vertical gradient in specific humidity in the

lower stratosphere is stronger in the BASCOE analyses. The

Southern Hemisphere polar vortex is drier in the BASCOE

analyses. Above 1 hPa the zonal mean specific humidity

fields vary quite considerably between the two systems. In

-90 -60 -30 0 30 60 90 Latitude

1000.0 100.0 10.0 1.0 0.1

Pressure, hPa

-90 -60 -30 0 30 60 90

Latitude 1000.0

100.0 10.0 1.0 0.1

Pressure, hPa

2 4 6 8

Specific Humidity ppmv

Fig. 1. Monthly zonal mean specific humidity analyses for September 2003 for BASCOE (upper plot) and ECMWF (lower plot). MIPAS water vapour profiles have been assimilated in both cases. Blue denotes relatively low specific humidity values; red denotes relatively high specific humidity values. Units: ppmv.

this region, the ECMWF analyses are ∼2 ppmv moister than the BASCOE analyses, which appear more realistic when compared to the UARS reference atmosphere. BASCOE analyses are ∼5% lower than MIPAS data in the lower meso- sphere, but the corresponding ECMWF analyses are ∼10%

greater. However, the ECMWF analyses are 25–30% too low compared to the uppermost MIPAS layer at 0.2–0.1 hPa. It appears that the ECMWF analyses aim to find a compromise between these conflicting biases in the MIPAS data, as we might expect given that a vertical smoothing is imposed by the background error correlations. Most of the differences between the two analyses can be explained by the fact that BASCOE does not assimilate any MIPAS data below 95 hPa and above 0.2 hPa (data outside these regions is model gen- erated). Influences from the troposphere and mesosphere are therefore excluded. The lack of any horizontal error correla- tions in the BASCOE assimilation scheme appears not be a problem because the MIPAS daily coverage is comparable to the BASCOE horizontal resolution.

The Met Office has investigated the impact of varying the control variable in the assimilation of MIPAS humidity data.

The objective is to develop a humidity control variable that has the desirable properties that it is usable in both the tro- posphere and the stratosphere, it has approximately Gaussian background errors, that temperature and humidity increments (i.e., the information that is added to the background field to produce the temperature and humidity analyses, respec- tively) are decoupled, and that allows realistic vertical error correlations. To achieve this, the Met Office have combined the ideas of Dee and da Silva (2003) and H´olm et al. (2002), and defined a normalized relative humidity variable. The im- pact of the normalization is to produce a considerably better conditioned background error covariance matrix and conse- quently the minimization in the 3D-Var algorithm is much faster. The removal of the influence of temperature incre- ments reduces spurious upper stratospheric increments.

Figure 2 compares mean MIPAS and Met Office anal-

ysed profiles of specific humidity for the 60 ^◦ S–90 ^◦ S region

on 25 September 2002. This is five days after the start of

Fig. 2. Met Office mean (left-hand plot) and standard deviation (right-hand plot) of specific humidity profiles for 25 September 2002 over the 60 ^◦ S–90 ^◦ S region. Black: MIPAS observations;

red: analyses using a relative humidity (RH) control variable; blue:

analyses using a normalized RH control variable; green: analyses using a normalized specific humidity control variable. Units: ppmv.

the assimilation experiment. Three different experiments are shown where the humidity control variable is either relative humidity (RH), normalized RH or normalized specific hu- midity. All three experiments show fairly reasonable specific humidity profiles below 5 hPa. However, at higher levels the fit to the MIPAS observations is less good, with the analy- ses being consistently too dry (see above for a discussion of the accuracy of MIPAS humidity data). The experiment with the normalized specific humidity control variable has a more reasonable lower mesospheric specific humidity, but is still too dry when compared to the MIPAS observations.

The reasons for the Met Office’s poor upper stratosphere and lower mesosphere assimilation are not fully understood.

It is possible that unrealistic vertical correlations in the back- ground error covariance matrix in the upper stratosphere and lower mesosphere may explain the poor specific humidity profiles in this region and their excessive variability. The normalized specific humidity control variable experiment, where the influence of temperature increments on the control variable is removed, has profiles with a reduced and more realistic standard deviation than the other two trials in the upper stratosphere, especially at around 4 hPa. The improve- ment in tropopause specific humidity seen in the two trials with the normalized control variable (RH or specific humid- ity) may result from the normalized variable handling better the steep gradient in specific humidity across the tropopause.

However, this improvement is not seen at all latitudes.

Although the dry bias in the Met Office upper strato- sphere/lower mesosphere cannot be explained by biases in the background or MIPAS observations at these levels, it was discovered that there are spurious correlations in the back- ground error covariances that can give rise to large humid-

ity increments at these upper levels from the assimilation of tropospheric observations. It is possible that a bias be- tween observations and background in the troposphere could erroneously be giving rise to a bias in the upper strato- sphere/lower mesosphere.

This section has given a brief overview of the humidity assimilation activities performed and results gained as part of the ASSET project. Evidence of a positive impact of hu- midity assimilation into DA systems (both operational and research) is provided. A more detailed intercomparison of the different humidity analyses is currently underway.

2.1.2 Assimilation of temperature retrievals

In this sub-section, results from experiments to assimilate Envisat temperature retrievals are shown. Evaluation stud- ies have shown that GOMOS temperature retrievals have large biases and are not ready to be validated (Goutail and Bazureau, 2004), while icing on the instrument has made SCIAMACHY temperature retrievals impossible (Piters et al., 2006). Therefore the temperature assimilation exper- iments were carried out using height-resolved retrievals (v4.61) from MIPAS only. The observations have been as- similated using both the ECMWF and Met Office DA sys- tems used in the ASSIC project (see Table 1).

MIPAS temperature profiles (v4.61) were monitored pas- sively in the ECMWF system by Dethof et al. (2004). The study indicates a bias of between –1 K and –6 K in the meso- sphere and near the stratopause, when compared with Met Office or ECMWF analyses, and between –2 K and 0 K when compared to National Centers for Environmental Prediction (NCEP) analyses. In the upper stratosphere, the MIPAS tem- peratures show a bias of between –2 K (compared to Met Of- fice analyses) and +4 K (compared to NCEP and ECMWF analyses). In the lower stratosphere the MIPAS tempera- tures have a positive bias of 0.5–2.5 K compared to the other datasets.

ECMWF ran a set of experiments for a 43-day period dur- ing August–September 2003. The control experiment assim- ilates all available operational data. The operational data for ECMWF encompassed a comprehensive blend of conven- tional (e.g. radiosondes, aircraft reports, profile data and sur- face weather stations) and satellite observations. The satel- lite data included: clear-sky radiances from four AMSU- A (Advanced Microwave Sounding Unit) instruments (Na- tional Oceanic and Atmospheric Administration, NOAA-15, -16 and -17, and EOS Aqua), AIRS (Atmospheric Infrared Sounder), AMSU-B, SSMI (Special Sensor Microwave Im- ager) and four geostationary satellites; Atmospheric Mo- tion Vectors (AMVs) from geostationary and polar satel- lites; scatterometer data; and radio occultation bending an- gle information from CHAMP (Challenging Minisatellite Payload). Ozone retrievals from SBUV (Solar Backscatter Ultraviolet Sounder) on NOAA-16 were also assimilated.

The test experiment assimilated the operational data plus

Fig. 3. Left-hand plot: Bias between MIPAS temperature retrievals and the ECMWF background (solid) and analysis (dotted) for the test experiment with assimilation of MIPAS retrievals (black) and the control experiment without assimilation of MIPAS data (grey). Statistics cover the region 60 ^◦ S–90 ^◦ S for the period 1–29 September 2003. Middle plot: As for left-hand plot, but for the standard deviation of the background and analysis departures. Right-hand plot: Number of MIPAS observations assimilated. Units for left-hand and middle plots: K.

-60 to -30 (209 obs)

-10 -5 0 5 10

temp /K 10.0

1.0 0.1

Pressure /hPa

-30 to 30 (267 obs)

-10 -5 0 5 10

temp /K

30 to 60 (41 obs)

-10 -5 0 5 10

temp /K

Fig. 4. Mean Met Office minus HALOE analysed temperature for MIPAS temperature and ozone assimilation (green), MIPAS ozone assimilation (red), MIPAS temperature and ozone assimilation with “Canadian Quick” (CQ, see text) background error covariances (blue) and MIPAS ozone assimilation with CQ background error covariances (orange). Left-hand plot: 60 ^◦ S–30 ^◦ S; centre plot: 30 ^◦ S–30 ^◦ N, right-hand plot: 30 ^◦ N–60 ^◦ N. The numbers in brackets indicate the HALOE/analysis coincidences within each latitude bin. Units: K.

height-resolved MIPAS temperature, humidity and ozone retrievals. Details of the operational data assimilated at ECMWF can be found at, for example, http://www.ecmwf.

int/research/ifsdocs/CY28r1/index.html.

Comparison of the results against MIPAS observations shows that assimilating MIPAS data has little impact on the analysis below the mid stratosphere, but that for levels above

∼5 hPa the bias between analysis and MIPAS retrievals, and the standard deviation of the analysis departures, is reduced.

An example of this appears in Fig. 3, which shows the bias

and standard deviation for the region 60 ^◦ S–90 ^◦ S. The bias of both the background and analysis in the upper stratosphere reduces from 4–8 K in the control run to 2–4 K in the test run.

The standard deviation of the background departures also re- duces compared to the control experiment by ∼2 K in this re- gion. Such reductions appear at all latitudes outside the trop- ics, which suggest that the information introduced through MIPAS is retained in the system.

It can be shown that, if biases and covariances of all data

are correctly specified, addition of new independent data will

result in a closer fit of the analysis to those data, so the ad- dition of MIPAS data should, in principle, improve the anal- yses. This is not always the case (as is shown in Fig. 3 for levels below ∼5 hPa) as, for example, there could be incon- sistencies in the assimilation system, for instance in the treat- ment of biases, etc.

Because comparison against independent data is much more significant than comparison against the assimilated observations, the ECMWF analyses were also compared against independent temperature data from HALOE (Hervig et al., 1996). This comparison (Fig. 4) shows that the in- formation introduced through MIPAS has a strong impact on the system. The bias of the analysis against HALOE, and its standard deviation, is reduced for levels above ∼5 hPa when MIPAS data are assimilated, although the reduction in stan- dard deviation is only a few tenths of degree K. These results have the caveats that the number of available HALOE pro- files for the comparison is much smaller than for MIPAS and that at levels below ∼6 hPa the HALOE temperature data are not retrievals but NCEP analyses.

MIPAS assimilation experiments were also run with the Met Office 3D-Var system for August 2003 (here, no MIPAS humidity data were assimilated). In contrast to the ECMWF results, the assimilation of MIPAS temperature results in a slight degradation in the Met Office temperature analyses.

This can be seen in Fig. 4, where the analyses from this experiment and a similar one in which no MIPAS temper- atures are assimilated, are compared with HALOE data. The bias with respect to HALOE generally increases in the up- per stratosphere and mesosphere with MIPAS temperature assimilation, especially in the Northern Hemisphere mid lat- itudes. Standard deviations of the analyses departures (not shown) are also larger.

A clue as to why this degradation takes place in the Met Office analyses comes from examination of opera- tional Met Office temperature analyses between August and November 2003. In October 2003, the Met Office forecast model changed from an Eulerian dynamical core to a semi- Lagrangian one (Davies et al., 2005). It was necessary to change the background error covariances when the dynam- ical core was changed. After the change, a clear degrada- tion appears in the analysis profile near 10 hPa, in the form of a jagged cold bias, together with a slight increase in the jaggedness of the profile at other levels. This suggests that the background error covariance used may be the source of the problem. The covariances are generated using the NMC method (Parrish and Derber, 1992), but for the new model the variances had to be constrained above the 10 hPa level in order to make the assimilation system more robust. In ret- rospect, it appears that the way in which this was done may have introduced noise into the background error covariances near the 10 hPa level.

This noisiness may explain the degradation in the analyses when MIPAS temperatures are used, since jaggedness in the individual profiles mentioned above is maintained or some-

times worsened when MIPAS temperatures are assimilated.

To test this, another pair of experiments was performed using background error covariances calculated using the method described by Polavarapu et al. (2005) (the so-called Cana- dian Quick, CQ, covariances). The CQ covariances are much smoother than the NMC covariances. Results from experi- ments with and without MIPAS temperatures and using the CQ covariances are also shown in Fig. 4. For levels above 10 hPa the biases for the CQ runs are smaller than for the corresponding NMC runs. In addition, with CQ covariances, the impact of the MIPAS temperature assimilation is neutral or slightly positive at almost all locations and the clear nega- tive impacts seen in Northern Hemisphere mid latitudes with the NMC covariances have been removed.

In summary, the results from ECMWF indicate that as- similation of MIPAS temperature retrievals helps to reduce biases in the temperature analyses for levels above ∼5 hPa, whereas little impact is found for levels below ∼10 hPa. This seems reasonable, since at upper levels the analysis is less well constrained from radiosondes or satellite radiance ob- servations than for levels below ∼10 hPa, and the bias cor- rection of the higher-peaking satellite channels is also less well characterized than at lower levels. The results from the Met Office underline the importance of using appropri- ate background error covariances in a DA system. When po- tential problems were identified and addressed, by replacing the NMC covariances with CQ covariances, the degradation caused by the assimilation of MIPAS temperatures vanished.

The absence of the positive impact seen at upper levels in the ECMWF analyses may perhaps be related to further differ- ences in background error covariances at these levels in the two systems, or perhaps different biases in the background fields.

2.1.3 Limb radiance assimilation

Direct radiance assimilation has been very successful for the

assimilation of temperature and humidity information from

nadir sounding instruments, and it is therefore used oper-

ationally at most major NWP centres (e.g. Saunders et al.,

1999; McNally et al., 2006). Prompted by this success, the

ASSET project applied for the first time the radiance assimi-

lation framework to the assimilation of limb radiances. Vari-

ational data assimilation schemes allow the direct assimila-

tion of radiances, without the additional step of performing

retrievals off-line (e.g. Andersson et al., 1994). The main

advantages are that radiance assimilation provides a fairly

generic framework for using radiances together with all other

observations and the latest background data, such that the

analysis/retrieval problem is better constrained. Radiance as-

similation also avoids the need to account for complex error

characteristics in retrievals, arising, for instance, through the

use of a short-term forecast as a priori, or from other as-

sumptions in the retrieval algorithm. At the same time, a

challenge commonly encountered in radiance assimilation is

that model-simulated and observed radiances almost always show systematic deviations or biases (e.g. Harris and Kelly, 2001). While some of these may be due to biases in the model fields, a large proportion is usually attributed to so- called “radiance biases”, i.e., biases arising from errors in the instrument characterization (e.g. spectral response func- tions), the calibration, the spectroscopy, or other aspects of the forward model. As DA schemes assume unbiased data, such biases need to be removed. The development of suitable ways to account for these biases is an area of active research in the case of nadir radiance assimilation (e.g. Dee, 2004).

Assimilation of limb radiances was applied to clear-sky emitted infrared radiances from MIPAS in the ECMWF sys- tem (e.g. Rabier et al., 2000). This required the development of a fast radiative transfer model, its tangent linear and ad- joint (Bormann et al., 2005), the selection of a suitable sub- set of MIPAS data to be used in the assimilation (Bormann and Healy, 2006), and the modification of the ECMWF sys- tem to be able to deal with the limb-viewing geometry. The study also touched on a number of other novel and exper- imental aspects, such as the extraction of ozone informa- tion through direct assimilation of ozone-affected radiances, and first experiences with combined tropospheric and strato- spheric humidity analyses, made possible through work by H´olm et al. (2002). The developments for the radiance as- similation also allow continuous monitoring of MIPAS radi- ances against model equivalents, opening, for example, new possibilities of characterizing the temporal stability of the in- strument. Results of the assimilation of MIPAS radiances are described in detail in Bormann and Th´epaut (2007) and Bor- mann et al. (2007); here we only give a brief overview of the main findings.

First trials over a period of 43 days in August/September 2003 demonstrate the feasibility of assimilating information from the MIPAS instrument by means of direct assimila- tion of radiances. In these experiments we use an observa- tion operator which assumes local horizontal homogeneity for the radiative transfer calculations, as is done in ESA’s routine retrieval processing (Ridolfi et al., 2000). Subse- quent experimentation relaxed this assumption, as further de- scribed below. We assimilate clear-sky radiances from 260 selected pseudo-channels over channel-specific tangent al- titude ranges. Overall, the assimilation is drawing well to the limb radiances, without significantly degrading the fit to other observations. This is an important result as it suggests that the MIPAS data and our assimilation approach are con- sistent with the rest of the observing network and its use in the DA system.

The assimilation has a considerable impact on the mean temperature, humidity, and ozone analyses in the strato- sphere, upper troposphere, and lower mesosphere (Fig. 5).

For instance, mean differences between analyses with and without MIPAS radiances show oscillating structures in the vertical, especially over the higher latitudes. Independent HALOE retrievals agree well with the features present in

the assimilation with MIPAS radiances (but the HALOE data provide only limited coverage of 60 ^◦ N–71 ^◦ N; Equator–

45 ^◦ S for the period in question). The radiance assimila- tion corrects erroneous temperature oscillations in the anal- yses; such oscillations are often referred to as “stratospheric ringing”, and they are a common problem in stratospheric data assimilation. Also, comparison with HALOE, SAGE II or POAM III data suggests that the radiance assimilation corrects a dry bias otherwise present throughout the strato- sphere (Fig. 6a). For ozone, results are more mixed. Com- parisons to independent data indicate improved ozone fields over the 60 ^◦ N–90 ^◦ N region (Fig. 6b), but with mean incre- ments which are too broad in the vertical over the tropics (not shown).

Our results show considerable sensitivity to how the cor- rection of biases in MIPAS radiances is handled. The pres- ence of radiance biases in MIPAS data becomes apparent when MIPAS radiances are assimilated without bias correc- tion. In this case, inconsistencies occur in the analysis biases for some MIPAS radiances whose weighting functions peak at similar altitudes, suggesting that these biases cannot be ac- counted for by biases in the model fields. The development of a bias correction for MIPAS is hampered by the presence of biases in the model fields. This makes it inappropriate to assume that the model fields without MIPAS data are unbi- ased when deriving the bias correction, as is commonly done for nadir radiances (Harris and Kelly, 2001). This is espe- cially true for channels sensitive to humidity or ozone in the stratosphere. The situation is very different to the case of nadir radiances sensitive to temperature over the much bet- ter observed troposphere, for which in situ data such as ra- diosondes and the presence of many different observations make cross-calibration of biases more feasible. Current ex- perimentation with MIPAS radiances uses the so-called γ /δ- method which scales optical depths in the radiative transfer model with a channel-specific γ , and models the remaining bias with a constant δ (Watts and McNally, 2004). More de- tails on the derivation of the bias correction and its influence can be found in Bormann et al. (2007).

The assimilation of MIPAS radiances has also been ex- tended to the use of a 2-D radiative transfer model (Bormann and Healy, 2006) which takes into account the horizontal gra- dients within the limb viewing plane (Bormann et al., 2007).

Our studies show that the assumption of horizontal homo- geneity can introduce a considerable forward model error.

Use of a 2-D radiative transfer model leads to smaller “First Guess” departures for these radiances. The smaller “First Guess” departures translate into smaller analysis increments, especially for ozone and humidity around the polar vortices.

However, statistics against independent observations show only relatively small improvements compared to using a 1- D radiative transfer model, mainly in the UTLS region.

The developments for the MIPAS limb radiance assim-

ilation allow a range of different aspects to be studied in

greater detail, and some of these will be further described

80N 60N 40N 20N 0 20S 40S 60S 80S 300

200 100 80 60 50 40 30 20 10 8 6 5 4 3 2 1 0.8 0.6 0.5 0.4 0.3 0.2

-16 16 16

32

32

32 32

48

48

48

80N 60N 40N 20N 0 20S 40S 60S 80S 300

200 100 80 60 50 40 30 20 10 8 6 5 4 3 2 1 0.8 0.6 0.5 0.4 0.3 0.2

0.2

0.2 0.4 0.4

80N 60N 40N 20N 0 20S 40S 60S 80S 300

200 100 80 60 50 40 30 20 10 8 6 5 4 3 2 1 0.8 0.6 0.5 0.4 0.3 0.2

-2

-1 -1 -1

1 1 1

1

3

4 5

Pressure [hPa]

(a) (b) (c)

Latitude [deg] Latitude [deg] Latitude [deg]

Fig. 5. (a) Zonal mean temperature differences between the ECMWF experiments with and without assimilation of MIPAS radiances.

Contour interval is 0.5 K, with positive values shown by solid black contour lines and negative values shown by dashed grey lines. (b) Same as (a), but for humidity (relative to the experiment without MIPAS radiance assimilation), with a contour interval of 8%. (c) Same as (a), but for ozone volume mixing ratio with a contour interval of 0.1 ppmv. Positive values indicate that the experiment with assimilation of MIPAS radiances has higher values than that without assimilation of MIPAS radiances. Units: (a) K; (b) percent; (c) ppmv.

-50 -40 -30 -20 -10 0 10 20 30 40 50

Humidity difference (%)

200 100 8070 60 50 40 30 20 10 87 6 5 4 3 2

Pressure (hPa)

-30 -25 -20 -15 -10 -5 0 5 10 15 20 25 30

Ozone difference (%)

200 100 8070 60 50 40 30 20 10 87 6 5 4 3 2

Pressure (hPa)

Fig. 6. Comparison of ECMWF analyses for the period 1–29 September 2003 against 195 SAGE II retrievals over the 60 ^◦ N–90 ^◦ N region.

Solid lines show the retrieval minus analysis difference (positive differences indicate that the analyses have a negative bias against SAGE II data); dotted lines show the standard deviation of the differences between the retrieval and the analysis, both relative to the mean retrieval.

Red lines indicate the experiment without assimilation of MIPAS data, black lines show statistics with MIPAS radiance assimilation. (a) Statistics for humidity analyses. (b) Statistics for ozone analyses. Units: percent.

in upcoming papers. This includes a comparison of assimi- lating retrievals versus assimilating radiances. Also, the de- velopments could be adapted for other passive limb-sounding instruments, such as the Microwave Limb Sounder (MLS) on the EOS Aura satellite.

2.2 Stratospheric distribution of chemical species

We focus on the nitrogen oxides NO _x and NO _y , which are

of primary importance in controlling stratospheric ozone

amounts. In the middle stratosphere, reactions involving

NO _x and NO _y form the primary catalytic O ₃ destruction

cycle. In the lower stratosphere, NO _x radicals moderate

O ₃ destruction by combining with hydrogen (HO _x ) and

halogen (ClO _x , BrO _x ) radicals involved in catalytic ozone destruction. It is therefore important to understand quan- titatively NO _x chemistry in order to estimate the chemical ozone budget and assess the impact of NO _x perturbations from, e.g., aircraft emissions or increasing N ₂ O emissions on stratospheric ozone levels. An additional motivation for assimilating short-lived species such as NO _x is the strong temperature-dependence of their chemistry. In the same way as the variability of tracer fields can provide information on winds, the variability of short-lived species could provide in- formation on temperature. One possible application is the use of temperature as a control variable in a chemical DA system. Note that, up to now, the ability to extract tempera- ture information from chemical measurements has not been tested.

The assimilation of short-lived chemical species (chem- istry timescales ≪ transport timescales) such as NO _x is more challenging than the assimilation of chemical tracers (chem- istry timescales ≫ transport timescales). This is because concentrations of short-lived species vary on timescales from less than a minute to one day, and hence detailed treatment of fast chemistry is required for simulating this variability.

This added level of complexity partly explains why DA sys- tems for short-lived chemical species are less common than for tracers.

We present summaries of two case studies that illustrate work done on the assimilation of NO x data within ASSET;

a subsequent publication will describe an intercomparison of NO 2 analyses from CTMs participating in the ASSET project. The first study is a test of NO x chemistry using a 4D-var photochemical assimilation system (Marchand et al., 2003). The second study concerns preliminary results in the extraction of temperature information from NO _x ob- servations. Both studies use height-resolved constituent data from the GOMOS instrument on board Envisat. GOMOS is a stellar occultation instrument making mostly night-time measurements of NO ₂ , O ₃ and, for the first time, NO ₃ (Marc- hand et al., 2004). The nitrate radical (NO ₃ ) is an important intermediate in the establishment of the partitioning between the NO _x and NO _y reservoir species. The NO _x night-time chemistry is understood to be relatively simple in the strato- sphere, with NO 3 concentrations controlled by temperature, and by the NO 2 and O 3 concentrations.

In the first study, GOMOS chemical data are assimilated into a CTM forced with ECMWF analyses. The objective is to test our understanding of NO _x night-time chemistry. The stratospheric photochemical scheme is standard and takes into account heterogeneous chemistry on sulphuric acid par- ticles assuming a background aerosol loading. Rate constants for the photochemical reactions are as in Sander et al. (2003).

The photochemical model is described in detail in Khattatov et al. (1999) and Marchand et al. (2003, 2004).

The CTM is coupled to a 4D-Var scheme that assumes (as is usual so far for 4D-Var) that the model is perfect, i.e., it has no errors associated with its temporal evolution. Further-

Fig. 7. GOMOS NO ₃ measurements as a function of analysed NO ₃ . Two periods are considered: (diamonds) 5–6 December 2002, and (filled circles) 30 January–2 February 2003. NO ₃ measurement ran- dom errors are indicated by the vertical lines. The linear regressions for the different periods are given by the dashed lines. The 1-1 line is represented by the thin black line. Units: ppbv. Based on Marc- hand et al. (2004).

more, to keep the experimental set up simple but capable of producing results that show the essence of the method, we also neglect forecast errors arising from errors in initial con- ditions. Thus, the only error term retained is that from the observations.

O ₃ and NO ₂ measurements from GOMOS are assimi- lated simultaneously. In all cases studied, the analysed O ₃ and NO ₂ fields match the corresponding O ₃ and NO ₂ GO- MOS measurements within 8.4×10 ⁻⁵ % and 0.0093%, re- spectively. The analysed NO ₃ field (i.e., NO ₃ calculated by the model after assimilation of GOMOS NO ₂ and O ₃ data) is then compared to the corresponding GOMOS NO ₃ measure- ment. A correlation plot of GOMOS NO ₃ versus analysed NO 3 is shown for 296 tropical cases in Fig. 7. The linear re- gression slope is 0.98 (±0.04) which is not statistically dif- ferent from the 1-1 line. Since NO 3 , NO 2 and O 3 are known to be strongly coupled chemically, this good agreement sug- gests that the NO _x chemistry scheme in the photochemical model is essentially correct and that this set of GOMOS mea- surements is chemically self-consistent. More details can be found in Marchand et al. (2004).

CTMs calculate the chemical composition of the strato-

sphere from chemical rate constants, and from off-line winds

and temperatures from meteorological analyses. It is inter-

esting to see to what extent the reverse procedure is possi-

ble, i.e., deriving winds and temperatures from the chemical

composition of the stratosphere. Some fast chemistry pro-

cesses are very sensitive to temperature changes, e.g., NO ₃

night-time chemistry. For a 1 K change, the NO ₃ concentra-

tion typically changes by more than 10%. This suggests that

temperature information could be inferred from changes in

the NO ₃ concentration.

Fig. 8. Mean difference (crosses) between GOMOS-derived tem- perature and ECMWF temperature as a function of (top) altitude in km, and (bottom) latitude in degrees. The standard deviations (dot- ted vertical lines) are also provided. To aid visualization of these diagrams, 3 K standard deviations (horizontal bars) are also plotted.

The GOMOS data is for the first four months of 2004. Units: K.

In the second study, we derive temperature information di- rectly from night-time NO 3 and O 3 concentrations assuming steady-state conditions for NO 3 . This steady-state hypothe- sis is generally valid in the lower and mid stratosphere. Using recommended values for the chemical rate constants, we de- rive temperature fields and evaluate them against ECMWF temperatures. The biases between these temperatures and GOMOS-derived temperatures are found to be less than 5 K throughout most of the lower and mid stratosphere. It is also possible to eliminate most of the bias by varying slightly the reaction rate constants within the errors reported for the rec- ommended kinetic data (Sander et al., 2003). Figure 8 shows an example for several months of GOMOS data in 2003, after adjustment of the chemical rate constants. Generally, the biases are very small and the root-mean-squared (RMS) statistic of the biases ranges between 2 K and 6 K depending

Logan/Fortuin/Kelder climatology MIMOSA

Juckes

MOCAGE-PALM Reprobus MOCAGE-PALM Cariolle v2.1 BASCOE v3q33

BASCOE v3d24 KNMI TEMIS

DARC/Met Office UM ECMWF MIPAS ECMWF operational

Fig. 9. Colour key used in Figs. 10–11.

on the star and period considered. The brighter the star, the smaller is the error in the GOMOS measurements and, thus, the smaller is the RMS statistic of the temperature bias. This result suggests that NO x data contain useful temperature in- formation.

2.3 Objective evaluation of ozone analyses

Given the attention ozone has received over the past decade, both for NWP and for studies of chemical distributions in the stratosphere (Rood, 2005), ASSET undertook an ozone in- tercomparison project (ASSIC) to provide an objective eval- uation of ozone analyses produced using different DA sys- tems and techniques. Most systems assimilated a common ozone observational dataset, i.e., MIPAS, though some as- similated SCIAMACHY. All things being equal, we would expect analyses incorporating good ozone data to perform better than those incorporating poor ozone data.

The resulting analyses were evaluated by comparison

against each other, ozone analyses from outside the project,

and independent observations (i.e., not assimilated). In this

section, we describe the highlights of the ASSIC project and

put them into context. Further details can be found in Geer

et al. (2006). One strong motivation for the ASSIC project is

that, by confronting these various DA systems and DA tech-

niques with the newly available Envisat observations, it is

possible both to gain an understanding of their strengths and

weaknesses, and to make new developments. Such an in-

tercomparison also provides insight into ozone assimilation

strategies, which we discuss later. There have been a num-

ber of previous intercomparisons between the ozone fields in

different CTMs (e.g. Bregman et al., 2001; Roelofs et al.,

2003), but to our knowledge the ASSIC project is the first

time ozone analyses have been compared.

-90 to -60 (237 obs)

-40 -20 0 20 40

% 100

10 1

Pressure /hPa

-60 to -30 (242 obs)

-40 -20 0 20 40

%

-30 to 30 (450 obs)

-40 -20 0 20 40

%

30 to 60 (355 obs)

-40 -20 0 20 40

%

60 to 90 (249 obs)

-40 -20 0 20 40

%

Fig. 10. Mean of analysis minus HALOE differences, normalized by climatology, for the period 18 August–30 November 2003. See Fig. 9 for colour key. The numbers in brackets indicate the HALOE/analysis coincidences within each latitude bin. Units: percent. These data are used to evaluate the performance of the ozone analyses. Based on Geer et al. (2006).

11 21 31 10 20 30 09 19 29 09 19 29 08 18 28

Aug 2003 Sep 2003 Oct 2003 Nov 2003

0 1 2 3 4 5

ozone /ppmm

Fig. 11. Comparison of analyses and ozonesondes at the South Pole, 68 hPa over the period August–November 2003. See Fig. 9 for colour key. Units: ppmm. Based on Geer et al. (2006).

The ASSIC project involved eleven sets of analyses from seven different DA systems (two systems based on NWP GCMs, and five systems based on CTMs). The NWP models did not include feedback between the analysed ozone field and the radiation field. The DA systems are summarized in Table 1. Figure 9 provides a colour key of the different sys- tems whose analyses are evaluated in Figs. 10–11. If ozone chemistry is included in the assimilation system, it is done either by highly detailed photochemical schemes, or via a parametrization, often known as a Cariolle scheme (e.g. Car- iolle and D´equ´e, 1986). A Cariolle scheme is a linearization

of ozone photochemistry around an equilibrium state, using parameters derived from a more detailed chemical model.

Most of the analyses focus on the stratosphere, but the scope of the ASSIC project spans from the troposphere to the mesosphere (Sect. 2.4 specifically discusses the assim- ilation of tropospheric constituents from Envisat). Analy- ses are interpolated from their native resolution onto a com- mon grid and then compared to independent ozone data from HALOE (Russell et al., 1993; http://haloedata.larc.nasa.

gov), ozonesondes (Komhyr et al., 1995; Thompson et al.,

2003a, b) and TOMS (Total Ozone Mapping Spectrometer;

McPeters et al., 1998), and to MIPAS (Fischer and Oel- haf, 1996; ESA, 2004). The comparison was done on a common grid rather than in observation space for organi- zational reasons; Geer et al. (2006) show that the error in- curred in this approach is not significant. Most data, fig- ures, and code are publicly available via the project website (http://darc.nerc.ac.uk/asset/assic).

Analyses were compared for the period July–November 2003. This period was chosen because of the availability of high quality MIPAS ozone data (Geer et al., 2006; La- hoz et al., 2006; Raspollini et al., 2006). Statistics were built up from the difference between analyses and obser- vations (Analyses minus Observations, AmO, differences).

In some cases, the AmO differences involve observations assimilated (e.g. MIPAS ozone); in other cases they in- volve independent unassimilated observations (e.g. HALOE ozone). The AmO statistics were binned into the follow- ing regions: 90 ^◦ S–60 ^◦ S; 60 ^◦ S–30 ^◦ S; 30 ^◦ S–30 ^◦ N; 30 ^◦ N–

60 ^◦ N; 60 ^◦ N–90 ^◦ N. Statistics were binned monthly; also for the entire period 18 August–30 November 2003 (before 18 August 2003, the DARC analyses were not adequately spun up).

Because ozone amounts vary by several orders of mag- nitude throughout the atmosphere, the AmO statistics were normalized with respect to the climatology of Fortuin and Kelder (1998) in the stratosphere and Logan (1999) in the troposphere, and displayed as a percentage. In this way, all regions in the atmosphere are given approximately equal weight.

Figure 10 shows that, through most of the stratosphere (50–2 hPa) AmO biases are usually within ±10% compared to the HALOE instrument. Similar results are obtained against ozonesonde data for levels in the lower stratosphere, 100–10 hPa (not shown). Biases and standard deviations in the AmO differences are larger in the UTLS, in the tropo- sphere, the mesosphere, and the Antarctic ozone hole region.

In these regions, some analyses do substantially better than others, and this is mostly due to differences in the models.

At the tropical tropopause, many analyses show positive bi- ases and excessive structure in the ozone fields, likely due to shortcomings in assimilated tropical wind fields and a degra- dation in MIPAS data at these levels. In the troposphere (lev- els below 100 hPa) some analyses show quite substantial bi- ases compared to ozonesonde observations (not shown). No ozone profiles for levels below ∼400 hPa are assimilated in any of the analyses, and only one analysis is designed to assimilate tropospheric chemical species (MOCAGE-PALM Reprobus – Table 1).

In the Southern Hemisphere ozone hole, only the anal- yses which correctly model heterogeneous ozone depletion are able to reproduce the ozone destruction over the Pole (Fig. 11). These analyses are those using Cariolle scheme versions 1.2 and 2.1, which include a term to take account of heterogeneous chemistry (operational ECMWF system; op- erational ECMWF system with MIPAS; MOCAGE-PALM

Cariolle scheme – see Table 1), or those using comprehensive chemical schemes (MOCAGE-PALM Reprobus; and BAS- COE v3d24 and v3q33 – Table 1). There are two other points worth mentioning: First, ECMWF operational analyses only capture the full ozone depletion during October when they began assimilating MIPAS ozone data for the first time, the benefit coming from the relatively high vertical resolution of MIPAS, and the fact that before this only limited ozone data were assimilated (Table 1); Second, the original BASCOE analyses (v3d24) were found to perform poorly in the ozone hole in the intercomparison – the scheme was improved and the new analyses (v3q33) performed better, demonstrating the value of intercomparisons for identifying shortcomings.

Most of the analyses (except KNMI TEMIS, who assim- ilate SCIAMACHY total column ozone data) at this level show too high ozone in November compared to ozoneson- des (Fig. 11). This is thought to be due to the relatively broad resolution of MIPAS (and SCIAMACHY) ozone profile data, thus the analyses show an influence from the much higher ozone amounts at levels above, where the polar vortex has already broken down.

In the upper stratosphere and mesosphere (levels above 5 hPa) some implementations of linear ozone photochemistry schemes cause large biases. However, these are easily reme- died (Geer et al., 2007). The diurnal cycle of mesospheric ozone is not captured, except by the one system that includes a detailed treatment of mesospheric chemistry (BASCOE – Table 1).

The conclusions drawn from the ASSIC project are that, in general, with current DA systems, in regions of good data quality and coverage, similarly good ozone analyses are ob- tained regardless of the DA method (KF; 3D- and 4D-Var;

3D-FGAT; direct inversion), or the model (NWP or CTM).

This reflects the generally good quality of the MIPAS ozone observations. There were areas where some models per- formed better than others, and in general the improved per- formance could be explained by better modelling of transport and chemistry.

One way of understanding the behaviour of the ozone analyses is to compare ozone photochemical relaxation timescales with the timescale on which Envisat observa- tions are available. Ozone has photochemical relaxation timescales of order one month in the UTLS, and of order one day in the upper stratosphere; in the mesosphere it has a di- urnal cycle. Envisat provides observations with daily global coverage, but the revisit cycle at a particular location is of order days.

Thus, chemical biases can in principle be adjusted by the

DA system in the UTLS, and the chemistry representation

is not generally crucial (although the ozone hole is an ex-

ception due to the role of heterogeneous chemistry – see

Geer et al., 2006). In the upper stratosphere (10–1 hPa), as

chemical timescales become faster, the influence of the ozone

observations in a DA system can become increasingly lim-

ited, and the chemistry representation becomes increasingly

important. Thus, it may be argued that in this region it is better to exclude chemistry than to model it inappropriately.

However, this will only work where observational data cov- erage is very good. As an example, the Juckes and MI- MOSA systems, which do not incorporate chemistry, com- pare as well to independent data in the upper stratosphere as the BASCOE analyses, which incorporate a detailed chem- istry model. Many of the analyses that use the linearized Cariolle scheme approach to chemistry, such as those from DARC/Met Office and ECMWF, show much poorer agree- ment with independent data. These problems are mainly due to avoidable biases in the schemes, but there are also funda- mental limitations with the approach. Hence, in the presence of the generally good quality and coverage of the MIPAS ozone observations in the upper stratosphere, the assimila- tion systems with no chemistry at all can do better than those with linearized chemistry.

Analyses based on SCIAMACHY total column ozone show a similar performance to the MIPAS analyses. The KNMI TEMIS set up, which uses the basic assimilation system described in Segers et al. (2005a), produces reason- ably realistic ozone profile shapes and stratospheric variabil- ity. However, significant model biases were detected in the upper stratosphere, mainly resulting from erroneous coeffi- cients in the upper levels of the variant of the Cariolle scheme that they were using (Geer et al., 2006). Separately, SCIA- MACHY limb profiles (Segers et al., 2005b) were assimi- lated, although they were only available in quantity in Oc- tober and November during the intercomparison period. De- tails of the analyses incorporating SCIAMACHY limb ozone profiles are given in Geer et al. (2006). In brief, these anal- yses do not perform as well as those for MIPAS ozone pro- files; this is likely due to pointing errors in v1.6 of the SCIA- MACHY ozone profile data, which have been documented in several publications (Segers et al., 2005b; Brinksma et al., 2006; von Savigny et al., 2005a, b).

Since the work discussed in Sect. 2.3 was done (2005), the data versions for both MIPAS and SCIAMACHY have changed. In particular, the latest version of SCIAMACHY ozone limb profile data (v1.63) now has a smaller pointing error than v1.6, although it is not completely eliminated (in- formation provided at the 3rd Atmospheric Chemistry Val- idation for Envisat, ACVE-3, meeting held 3–7 December 2006). Thus, assimilation of these latest SCIAMACHY ozone limb profile data would be expected to give better ozone analyses than those described in Geer et al. (2006).

Using the analyses to compare MIPAS and independent observations indirectly, and treating MIPAS observations as point retrievals, we can evaluate MIPAS ozone data: it is

∼5% higher than HALOE ozone above 30 hPa, and ∼10%

higher than ozonesondes and HALOE ozone in the lower stratosphere (100–30 hPa). This indicates MIPAS ozone data have a positive bias of 5–10% in the stratosphere and meso- sphere against ozonesondes and HALOE data.

The ASSIC project also provides some clues on ozone as- similation strategies. The choice of system for ozone DA depends on a range of considerations, including history, fa- miliarity with a particular approach or cost. If a central concern is the production of weather forecasts, it makes sense to add ozone to enhance a pre-existing NWP system (e.g. ECMWF). Alternatively, if the main focus is on chem- istry, there are strong arguments to build ozone assimilation into a CTM with sophisticated chemistry, taking the mete- orological input as given (e.g. BIRA-IASB). If the focus is primarily on ozone itself, an alternative approach is to use a transport model, driven by pre-existing dynamical fields, in combination with a simplified chemistry scheme. An exam- ple is KNMI, who have developed semi-operational SCIA- MACHY ozone analyses and forecasts (http://www.temis.

nl).

The ASSIC results suggest that for current DA systems, provided there is good data quality and coverage, similar quality of ozone analyses is obtained whether one uses a DA system based on an NWP model or a CTM. However, the cost of the different DA systems can be an issue. For ex- ample, GCM-based analyses require substantially more com- puter power than the CTM approach, though ozone assimila- tion is a relatively small additional cost when included in an existing NWP system (see, e.g., Geer et al., 2006).

2.4 Distribution of tropospheric pollutants

As part of the ASSET project, the spatial and temporal dis- tribution of tropospheric pollutants was studied. A strong motivation is the need to monitor and forecast air quality, for example, as part of the EC/ESA Global Monitoring for the Enviroment and Security (GMES) initiative. Because the assimilation of tropospheric constituents provides an objec- tive framework for monitoring and forecasting tropospheric pollutants, one ASSET partner (University of K¨oln) studied the impact of assimilating tropospheric constituents from En- visat into a DA system based on the University of K¨oln EU- RAD (European Air Pollution Dispersion) CTM (Elbern and Schmidt, 2001). The results from this study are summarized in this section.

Most satellite instruments are currently unable to provide

height-resolved profiles of constituents below tropopause

levels. Furthermore, the presence of clouds often prevents

retrievals of tropospheric columns of constituents, even when

spectral characteristics of constituent absorption theoreti-

cally make this possible. For example, although MIPAS can

theoretically retrieve information down to heights of 7 km

in the absence of clouds (Spang et al., 2005), its ability to

sound the troposphere is severely impaired by the presence

of clouds. In the absence of clouds, MIPAS profiles (e.g. of

ozone and HNO ₃ ) are available for the upper half of the

troposphere, and SCIAMACHY provides integrated tropo-

spheric column information: e.g., global fields of NO ₂ and

SO ₂ (see http://www.temis.nl).